DetReIDX: A Stress-Test Dataset for Real-World UAV-Based
Person Recognition

¹University of Beira Interior    ²J.N.N. College of Engineering    ³SCS, SRTM University
⁴Istanbul Medipol Üniversitesi    ⁵SRMIST
arXiv GitHub Leaderboard Download Data
7 Universities  |  509 Identities  |  13M+ Annotations

DetReIDX: Long-range identification and recognition from UAV and ground cameras.


News

Workshop

1st Workshop on VReID-XFD: Video-based Human Recognition at Extreme Far Distances

6 March 2026    Tucson, Arizona, USA    WACV 2026

Organizer

We organized the 1st Workshop on Video-based Human Recognition at Extreme Far Distances (VReID-XFD) based on the DetReIDX dataset at WACV 2026, Tucson, Arizona.

The workshop brought together researchers advancing UAV-based person re-identification, long-range video analytics, and extreme-distance recognition. We thank all authors, reviewers, and competition participants for their contributions, with special thanks to Prof. Hugo Proença for leading the DetReIDX challenge.

Workshop papers: https://lnkd.in/eb-mKKiB

Paper

Paper accepted at IEEE Transactions on Biometrics, Behavior, and Identity Science

Hambarde et al., "DetReIDX: A Stress-Test Dataset for Real-World UAV-Based Person Recognition", 2026.

IEEE T-BIOM    2026


Abstract

We introduce DetReIDX, a new large-scale benchmark dataset designed for real-world, long-range human recognition. It supports four core computer vision tasks: person detection, re-identification (ReID), multi-view tracking, and action recognition — all captured in complex outdoor scenes using UAV drones and ground cameras across seven international collection sites.

DetReIDX contains 509 unique identities with over 13 million bounding box annotations, captured from 18 UAV viewpoints per subject at altitudes ranging from 5m to 120m and distances of 10–120m. Each subject wears different outfits across sessions to evaluate long-term clothing-change robustness. Every frame is labeled with 16 soft biometric attributes (age, gender, clothing, action, etc.), providing fine-grained annotations for comprehensive evaluation of human-centric AI under real-world aerial surveillance conditions.

DetReIDX Overview

Figure 1: Comparison between existing datasets (ground-ground, aerial-aerial, and aerial-ground) and DetReIDX. Unlike counterparts, DetReIDX includes clothing variation, detection and tracking annotations, action labels, and wide aerial altitude coverage (5.8m–120m), making it well-suited for long-range surveillance tasks.

509
Identities
13M+
Annotations
18
UAV Viewpoints
120m
Max Altitude
7
Collection Sites

Dataset Examples

Sample recordings from each participating institution.

Outdoor UAV Examples
Indoor Reference Session

Indoor profile and gait capture session: front, left, and right angles per subject.


Dataset

Comparison with Existing Datasets

DetReIDX exceeds prior datasets in altitude span, viewpoint coverage, identity diversity, and annotation richness. DetReIDX (row 19, highlighted) is the only dataset combining aerial altitudes up to 120m, cross-session clothing variation, and all five task annotations.

No. Dataset Camera View Format Det. Track. ReID Search Action PIDs BBox Height (m) Distance (m)
17AG-ReID.v2 [17]UAV + CCTVGround + AerialStill1615100.6K15~45-
18G2APS-ReID [18]UAV + CCTVGround + AerialStill2788200.8K20~60-
19DetReIDX (Ours)DSLR + UAVGround + AerialVideo + Still 33413M5~12010~120

Research Challenges

DetReIDX exposes critical challenges in person recognition that are overlooked in traditional datasets but common in real-world UAV surveillance:

Extreme Scale Variation

Person ROIs range from full-HD indoor captures to sub-10px silhouettes at 120m altitude, testing resolution robustness.

Clothing Variation

Subjects wear different outfits across sessions, requiring models to learn identity beyond superficial appearance cues.

Viewpoint Diversity

18 unique UAV perspectives across three pitch angles (30°, 60°, 90°) challenge current view-specific approaches.

Cross-Domain Transfer

Aerial-to-ground matching requires bridging vastly different capture modalities and perspectives.

Occlusion & Blur

Real-world interference from motion blur, atmospheric conditions, and partial visibility.

Temporal Drift

Multi-day sessions with environmental changes test long-term recognition capabilities.


Data Collection Protocol

Data was collected using high-resolution drones (DJI Phantom 4) and DSLR cameras through a multi-institutional collaboration across Portugal, Angola, Turkey, and India, under diverse altitudes (5–120m), distances (10–120m), and pitch angles (30°, 60°, 90°).

Setting Environment Altitude Distance Data Type
Indoor University Labs Ground level Close range Profile images, gait videos
Outdoor University Campuses 5–120m 10–120m Multi-view videos, action clips

Annotation Protocol

The dataset includes over 13 million manually annotated bounding boxes for 509 unique identities, created with CVAT and verified by multiple annotators. Each subject is annotated with 16 soft biometric attributes covering demographics, appearance, and activity.

# Attribute Values / Description
1Gender0: Male, 1: Female, 2: Unknown
2Age0-11, 12-17, 18-24, 25-34, 35-44, 45-54, 55-64, >65, Unknown
3Height0: Child, 1: Short, 2: Medium, 3: Tall, 4: Unknown
4Body Volume0: Thin, 1: Medium, 2: Fat, 3: Unknown
5Ethnicity0: White, 1: Black, 2: Asian, 3: Indian, 4: Unknown
6Hair Color0: Black, 1: Brown, 2: White, 3: Red, 4: Gray, 5: Occluded, 6: Unknown
7Hairstyle0: Bald, 1: Short, 2: Medium, 3: Long, 4: Horse Tail, 5: Unknown
8Beard0: Yes, 1: No, 2: Unknown
9Moustache0: Yes, 1: No, 2: Unknown
10Glasses0: Normal Glass, 1: Sun Glass, 2: No, 3: Unknown
11Head Accessories0: Hat, 1: Scarf, 2: Necklace, 3: Cannot see, 4: Unknown
12Upper Body Clothing0: T-Shirt, 1: Blouse, 2: Sweater, 3: Coat, 4: Bikini, 5: Naked, 6: Dress, 7: Uniform, 8: Shirt, 9: Suit, 10: Hoodie, 11: Cardigan, 12: Unknown
13Lower Body Clothing0: Jeans, 1: Leggings, 2: Pants, 3: Shorts, 4: Skirt, 5: Bikini, 6: Dress, 7: Uniform, 8: Suit, 9: Unknown
14Feet0: Sport Shoe, 1: Classic Shoe, 2: High Heels, 3: Boots, 4: Sandal, 5: Nothing, 6: Unknown
15Accessories0: Bag, 1: Backpack, 2: Rolling Bag, 3: Umbrella, 4: Sport Bag, 5: Market Bag, 6: Nothing, 7: Unknown
16Action0: Walking, 1: Running, 2: Standing, 3: Sitting, 4: Cycling, 5: Exercising, 6: Petting, 7: Talking on Phone, 8: Leaving Bag, 9: Fall, 10: Fighting, 11: Dating, 12: Offending, 13: Trading

Full experimental results (YOLOv8, DDOD, Grid-RCNN, PersonViT, SeCap, CLIP-ReID) are available in the paper.


Dataset Access

DetReIDX is available for academic and non-commercial research use.

Person Re-ID

Still + video ReID splits with attribute labels

Download
Person Detection

Bounding box annotations for UAV & ground views

Video ReID

Multi-session tracklet sequences across altitudes


Citation

Acknowledgements

We acknowledge and give credit to the following universities for their contributions: Istanbul Medipol University, J.N.N College of Engineering, SRM Institute of Science and Technology, Swami Ramanand Teerth Marathwada University Nanded, Universidade Beira Interior, Universidade de Luanda. (Sorted in A–Z order)

University Logos

BibTeX

@article{hambarde2026detreidx,
  title={Detreidx: A stress-test dataset for real-world uav-based person recognition},
  author={Hambarde, Kailash A and Mbongo, Nzakiese and Kumar, MP Pavan and Mekewad, Satish and Fernandes, Carolina and Silahtaro{\u{g}}lu, G{\"o}khan and Nithya, Alice and Wasnik, Pawan and Rashidunnabi, MD and Samale, Pranita and others},
  journal={IEEE Transactions on Biometrics, Behavior, and Identity Science},
  year={2026},
  publisher={IEEE}
}